This study primarily examines real estate transaction data covering the period from the first quarter of 2011 to the fourth quarter of 2024, using quarterly observations. Housing prices and transaction volume are obtained from the Real Estate Actual Transaction Price Database of Taiwan Region’s Ministry of the Interior. The empirical analysis focuses on six special municipalities—New Taipei City, Taipei City, Taoyuan City, Taichung City, Tainan City, and Kaohsiung City. Housing prices and transaction volume are treated as the endogenous variables. The definitions of the variables used in this study are summarized in Table 1.

Table 1.Variable description.

Notes: Housing prices (HP) are measured in NT$10,000 per unit, and transaction volume (TQ) is measured by the number of units. All data are obtained from the Real Estate Actual Transaction Price Database of the Ministry of the Interior, Taiwan Region.

Housing prices (HP) and transaction volume (TQ) are constructed for each of the six special municipalities using quarterly data from the Real Estate Actual Transaction Price Database of the Ministry of the Interior (MOI). Specifically, the housing price variable is measured as the average transaction price per residential unit for each municipality in each quarter (in NT$10,000 per ping), and is expressed in units of NT$10,000. The sample includes only residential property transactions, excluding non-residential types such as commercial or industrial properties.

To enhance comparability across cities and over time, the construction of the housing price variable follows a consistent definition and data filtering procedure for all municipalities. In addition, by aggregating transaction-level data into quarterly averages, short-term fluctuations due to compositional changes in housing types are mitigated, allowing for more reliable comparisons of housing price dynamics across regions and periods.

Table 2 reports the descriptive statistics of housing prices for the six special municipalities, while Table 3 presents the descriptive statistics of housing transaction volume.

Table 2.Descriptive statistics of housing prices in the six special municipalities.

Table 3.Descriptive statistics of housing transaction volume in the six special municipalities.

The Central Bank of Taiwan Region has implemented seven rounds of selective credit control measures since December 2020, aiming to curb speculative activities in the housing market. The key policy measures have focused on tightening loan-to-value (LTV) ratios for second and subsequent housing purchases, eliminating grace periods, and raising reserve requirement ratios.

Given that these policies may have induced structural changes in the relationship between housing prices and transaction volume, this study introduces a policy dummy variable. The dummy variable is set to 1 from the fourth quarter of 2020 onward, and 0 for periods prior to the third quarter of 2020. In this study, all variables are expressed in natural logarithms, except for those used in descriptive statistics, in order to smooth the data and stabilize variability.

Table 4 shows that, under the three Augmented Dickey-Fuller (ADF) test specifications, most variables are found to be non-stationary in levels. Although transaction volumes in New Taipei City, Taipei City, and Tainan City appear stationary under certain specifications, the results are not consistent across all testing models. Given that cointegration analysis and VECM (or VAR) estimation require variables to be integrated of the same order, the overall evidence supports treating all variables as non-stationary in levels. This ensures the validity and consistency of the subsequent empirical analysis.

After first differencing, all variables become stationary across specifications. Although the CPI series exhibits non-stationarity under certain specifications, the overall evidence supports treating all variables as stationary in first differences. Therefore, following standard practice and considering the overall evidence across different deterministic specifications, all variables are treated as integrated of order one, I(1), which justifies the use of the Johansen cointegration test and the VECM framework.

Table 4.ADF test.

Note: *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively.

To mitigate the effects of inflation, seasonal fluctuations, and credit policy interventions, several adjustments are incorporated into the cointegration analysis. First, nominal housing prices are adjusted for inflation to examine the relationship between real housing prices and transaction volume. Second, seasonal dummy variables are introduced to control for seasonal effects in housing prices and transaction activity. Third, a credit control policy dummy variable is included to account for the impact of macroprudential credit regulations on housing prices and transaction volume.

In the cointegration tests and VECM specification, the selection of deterministic components is based on information criteria. Specifically, alternative specifications with different combinations of intercepts and trends are evaluated, and the model with the lowest Akaike Information Criterion (AIC) is selected. The results indicate that the specification including both an intercept and a linear trend provides the best fit. This corresponds to a specification with an unrestricted intercept and linear trend, ensuring consistency between the Johansen test and the VECM estimation.

As reported in Table 5, the results of the Johansen cointegration tests indicate that the null hypothesis of no cointegration is rejected for all six municipalities based on the Trace test. The Maximum eigenvalue test also rejects the null hypothesis in most cases, although it fails to do so for Taipei City and Tainan City, and shows only marginal significance for Taoyuan City.

Despite minor inconsistencies between the two test statistics, the overall evidence supports the existence of a long-run cointegration relationship between housing prices and transaction volume across all six municipalities. Although the Trace test suggests the possibility of more than one cointegrating vector, the bivariate nature of the system implies that at most one cointegrating relationship can exist. Accordingly, the cointegration rank is restricted to one.

Based on these findings, housing prices and transaction volume are considered to be cointegrated in New Taipei City, Taipei City, Taoyuan City, Taichung City, Tainan City, and Kaohsiung City. Consequently, a Vector Error Correction Model (VECM) is employed for each municipality to capture both the long-run equilibrium relationship and the short-run dynamic adjustments between the two variables.

Table 5.Cointegration test results and model selection (VAR or VECM).

Note: Reported values are p-values. *, **, and *** denote statistical significance at the 10%, 5%, and 1% levels, respectively.

Table 6 reports the estimation results of the Vector Error Correction Model (VECM) for real housing prices and transaction volume across the six special municipalities, incorporating seasonal dummy variables and a credit control policy dummy variable.

The long-run cointegrating relationships indicate that transaction volume enters the real housing price equation significantly in most cities, confirming the existence of a long-run equilibrium relationship between real housing prices and transaction activity. Specifically, the cointegrating coefficients on transaction volume are positive in New Taipei City, Taipei City, Taichung City, Tainan City, and Kaohsiung City, while it is negative in Taoyuan City, reflecting regional heterogeneity in housing market structures.

Given that the cointegrating vector is normalized on real housing prices, the estimated relationship can be rearranged such that real housing prices depend on transaction volume. A positive coefficient on transaction volume therefore implies a negative long-run relationship, whereas a negative coefficient indicates a positive relationship. In New Taipei City, Taipei City, Taichung City, Tainan City, and Kaohsiung City real housing prices and transaction volume exhibit negative co-movement, suggesting that rising prices are not supported by an expansion in effective demand but are more closely associated with supply-side constraints such as land scarcity that limit transaction activity. By contrast, in Taoyuan City, the cointegrating coefficients of transaction volume are negative, suggesting that real housing prices and transaction volume move in the same direction.

With respect to short-run dynamics, the error correction term (CointEq1) exhibits heterogeneous adjustment patterns across municipalities. When the sign of housing quantity in cointegration vector is negative, a negative (positive) and statistically significant error correction coefficient in the housing price (transaction volume) equation indicates stable adjustment toward the long-run equilibrium, whereas a positive (negative) coefficient implies an unstable response to disequilibrium. When the sign of transaction volume in cointegration vector is positive, a negative (negative) and statistically significant error correction coefficient in the housing price (transaction volume) equation indicates stable adjustment toward the long-run equilibrium, whereas a positive (positive) coefficient implies an unstable response to disequilibrium.

In particular, for Taichung City, Tainan City, and Kaohsiung City, the error correction terms are negative and statistically significant in the transaction volume equations, indicating that transaction volume plays the primary role in restoring the long-run equilibrium following short-run shocks.

In New Taipei City and Taipei City, the error correction terms are negative and statistically significant in both the housing price and transaction volume equations, suggesting that neither variable exhibits a stable adjustment toward the long-run equilibrium.

By contrast, although the error correction term is statistically significant only in the housing price equation for Taoyuan City, indicating that only real housing prices respond to deviations from the long-run equilibrium, while transaction volume does not exhibit a significant adjustment. The negative sign of the adjustment coefficient suggests that real housing price does contribute to stabilizing the long-run equilibrium in those cases.

Overall, in terms of real housing prices, these results reveal substantial cross-city heterogeneity in long-run adjustment mechanisms, with stable error correction observed in most municipalities.

Furthermore, the coefficients on the lagged first-difference terms indicate significant short-run interactions between real housing prices and transaction volume. In Taipei City and Kaohsiung City, lagged changes in real housing prices exert a significant effect on transaction volume, implying a unidirectional influence from prices to volume. In contrast, in Taoyuan City, lagged changes in transaction volume significantly affect real housing prices in the short run.

In Taichung City, lagged changes in transaction volume significantly affect real housing price movements, while lagged real housing price changes also influence transaction volume, indicating bidirectional interactions between transaction volume and real housing prices. In New Taipei City and Tainan City, no significant short-run interaction between the two variables is found. Overall, these results suggest that short-run dynamics vary across regions.

Overall, the VECM results confirm that real housing prices and transaction volume in Taiwan’s six special municipalities are linked by both long-run equilibrium relationships and short-run adjustment mechanisms.

Table 6.VECM estimation results for real housing prices and transaction volume.

Note: t-statistics in ( ); *, **, and *** denote statistical significance at the 10%, 5%, and 1% levels, respectively.

In this subsection, we examine the long-run and short-run relationships between nominal housing prices and transaction volume to assess the robustness of the VECM estimation results. Table 7 reports the estimation results of the Vector Error Correction Model (VECM) for nominal housing prices and transaction volume across the six special municipalities, with seasonal dummy variables and a credit control policy dummy variable included in the specification.

Comparing Table 6 and Table 7, the long-run coefficients associated with transaction volume exhibit noticeable differences in magnitude between the real and nominal specifications. In particular, the effect of transaction volume becomes stronger in some municipalities (notably Taichung and Kaohsiung) when nominal prices are used, suggesting that nominal factors may amplify demand-driven price dynamics. Nevertheless, the signs of the cointegration coefficients remain unchanged across specifications, indicating that the long-run relationships are qualitatively robust.

Table 7.VECM estimation results for nominal housing prices and transaction volume.

Note: t-statistics in ( ); *, **, and *** denote statistical significance at the 10%, 5%, and 1% levels, respectively.

In the short run, the adjustment dynamics are broadly similar between the two models. One notable exception is New Taipei City, where the error correction term becomes statistically insignificant under the nominal specification. In addition, the absolute magnitude of the adjustment coefficients is generally smaller in the nominal model, implying a slower speed of convergence toward the long-run equilibrium.

Overall, these findings indicate that both the long-run relationships and short-run dynamics are largely robust to the choice between real and nominal housing prices. However, the adjustment process appears to be more efficient when housing prices are measured in real terms.